ARM Skia NEON patches - 19 - Intrinsics version of the Filter32 routines
BitmapProcState: translate the filtering routines to intrinsics
Signed-off-by: Kévin PETIT <kevin.petit@arm.com>
BUG=
R=djsollen@google.com, mtklein@google.com
Author: kevin.petit.arm@gmail.com
Review URL: https://chromiumcodereview.appspot.com/21915004
git-svn-id: http://skia.googlecode.com/svn/trunk/src@11246 2bbb7eff-a529-9590-31e7-b0007b416f81
diff --git a/opts/SkBitmapProcState_filter_neon.h b/opts/SkBitmapProcState_filter_neon.h
index 86c1dcf..e56b683 100644
--- a/opts/SkBitmapProcState_filter_neon.h
+++ b/opts/SkBitmapProcState_filter_neon.h
@@ -7,82 +7,86 @@
*/
+#include <arm_neon.h>
#include "SkColorPriv.h"
/*
- Filter_32_opaque
-
- There is no hard-n-fast rule that the filtering must produce
- exact results for the color components, but if the 4 incoming colors are
- all opaque, then the output color must also be opaque. Subsequent parts of
- the drawing pipeline may rely on this (e.g. which blitrow proc to use).
+ * Filter_32_opaque
+ *
+ * There is no hard-n-fast rule that the filtering must produce
+ * exact results for the color components, but if the 4 incoming colors are
+ * all opaque, then the output color must also be opaque. Subsequent parts of
+ * the drawing pipeline may rely on this (e.g. which blitrow proc to use).
*/
static inline void Filter_32_opaque_neon(unsigned x, unsigned y,
SkPMColor a00, SkPMColor a01,
SkPMColor a10, SkPMColor a11,
SkPMColor *dst) {
- asm volatile(
- "vdup.8 d0, %[y] \n\t" // duplicate y into d0
- "vmov.u8 d16, #16 \n\t" // set up constant in d16
- "vsub.u8 d1, d16, d0 \n\t" // d1 = 16-y
+ uint8x8_t vy, vconst16_8, v16_y, vres;
+ uint16x4_t vx, vconst16_16, v16_x, tmp;
+ uint32x2_t va0, va1;
+ uint16x8_t tmp1, tmp2;
- "vdup.32 d4, %[a00] \n\t" // duplicate a00 into d4
- "vdup.32 d5, %[a10] \n\t" // duplicate a10 into d5
- "vmov.32 d4[1], %[a01] \n\t" // set top of d4 to a01
- "vmov.32 d5[1], %[a11] \n\t" // set top of d5 to a11
+ vy = vdup_n_u8(y); // duplicate y into vy
+ vconst16_8 = vmov_n_u8(16); // set up constant in vconst16_8
+ v16_y = vsub_u8(vconst16_8, vy); // v16_y = 16-y
- "vmull.u8 q3, d4, d1 \n\t" // q3 = [a01|a00] * (16-y)
- "vmull.u8 q0, d5, d0 \n\t" // q0 = [a11|a10] * y
+ va0 = vdup_n_u32(a00); // duplicate a00
+ va1 = vdup_n_u32(a10); // duplicate a10
+ va0 = vset_lane_u32(a01, va0, 1); // set top to a01
+ va1 = vset_lane_u32(a11, va1, 1); // set top to a11
- "vdup.16 d5, %[x] \n\t" // duplicate x into d5
- "vmov.u16 d16, #16 \n\t" // set up constant in d16
- "vsub.u16 d3, d16, d5 \n\t" // d3 = 16-x
+ tmp1 = vmull_u8(vreinterpret_u8_u32(va0), v16_y); // tmp1 = [a01|a00] * (16-y)
+ tmp2 = vmull_u8(vreinterpret_u8_u32(va1), vy); // tmp2 = [a11|a10] * y
- "vmul.i16 d4, d7, d5 \n\t" // d4 = a01 * x
- "vmla.i16 d4, d1, d5 \n\t" // d4 += a11 * x
- "vmla.i16 d4, d6, d3 \n\t" // d4 += a00 * (16-x)
- "vmla.i16 d4, d0, d3 \n\t" // d4 += a10 * (16-x)
- "vshrn.i16 d0, q2, #8 \n\t" // shift down result by 8
- "vst1.32 {d0[0]}, [%[dst]] \n\t" // store result
- :
- : [x] "r" (x), [y] "r" (y), [a00] "r" (a00), [a01] "r" (a01), [a10] "r" (a10), [a11] "r" (a11), [dst] "r" (dst)
- : "cc", "memory", "d0", "d1", "d3", "d4", "d5", "d6", "d7", "d16"
- );
+ vx = vdup_n_u16(x); // duplicate x into vx
+ vconst16_16 = vmov_n_u16(16); // set up constant in vconst16_16
+ v16_x = vsub_u16(vconst16_16, vx); // v16_x = 16-x
+
+ tmp = vmul_u16(vget_high_u16(tmp1), vx); // tmp = a01 * x
+ tmp = vmla_u16(tmp, vget_high_u16(tmp2), vx); // tmp += a11 * x
+ tmp = vmla_u16(tmp, vget_low_u16(tmp1), v16_x); // tmp += a00 * (16-x)
+ tmp = vmla_u16(tmp, vget_low_u16(tmp2), v16_x); // tmp += a10 * (16-x)
+
+ vres = vshrn_n_u16(vcombine_u16(tmp, vcreate_u16(0)), 8); // shift down result by 8
+ vst1_lane_u32(dst, vreinterpret_u32_u8(vres), 0); // store result
}
static inline void Filter_32_alpha_neon(unsigned x, unsigned y,
SkPMColor a00, SkPMColor a01,
SkPMColor a10, SkPMColor a11,
SkPMColor *dst, uint16_t scale) {
- asm volatile(
- "vdup.8 d0, %[y] \n\t" // duplicate y into d0
- "vmov.u8 d16, #16 \n\t" // set up constant in d16
- "vsub.u8 d1, d16, d0 \n\t" // d1 = 16-y
+ uint8x8_t vy, vconst16_8, v16_y, vres;
+ uint16x4_t vx, vconst16_16, v16_x, tmp, vscale;
+ uint32x2_t va0, va1;
+ uint16x8_t tmp1, tmp2;
- "vdup.32 d4, %[a00] \n\t" // duplicate a00 into d4
- "vdup.32 d5, %[a10] \n\t" // duplicate a10 into d5
- "vmov.32 d4[1], %[a01] \n\t" // set top of d4 to a01
- "vmov.32 d5[1], %[a11] \n\t" // set top of d5 to a11
+ vy = vdup_n_u8(y); // duplicate y into vy
+ vconst16_8 = vmov_n_u8(16); // set up constant in vconst16_8
+ v16_y = vsub_u8(vconst16_8, vy); // v16_y = 16-y
- "vmull.u8 q3, d4, d1 \n\t" // q3 = [a01|a00] * (16-y)
- "vmull.u8 q0, d5, d0 \n\t" // q0 = [a11|a10] * y
+ va0 = vdup_n_u32(a00); // duplicate a00
+ va1 = vdup_n_u32(a10); // duplicate a10
+ va0 = vset_lane_u32(a01, va0, 1); // set top to a01
+ va1 = vset_lane_u32(a11, va1, 1); // set top to a11
- "vdup.16 d5, %[x] \n\t" // duplicate x into d5
- "vmov.u16 d16, #16 \n\t" // set up constant in d16
- "vsub.u16 d3, d16, d5 \n\t" // d3 = 16-x
+ tmp1 = vmull_u8(vreinterpret_u8_u32(va0), v16_y); // tmp1 = [a01|a00] * (16-y)
+ tmp2 = vmull_u8(vreinterpret_u8_u32(va1), vy); // tmp2 = [a11|a10] * y
- "vmul.i16 d4, d7, d5 \n\t" // d4 = a01 * x
- "vmla.i16 d4, d1, d5 \n\t" // d4 += a11 * x
- "vmla.i16 d4, d6, d3 \n\t" // d4 += a00 * (16-x)
- "vmla.i16 d4, d0, d3 \n\t" // d4 += a10 * (16-x)
- "vdup.16 d3, %[scale] \n\t" // duplicate scale into d3
- "vshr.u16 d4, d4, #8 \n\t" // shift down result by 8
- "vmul.i16 d4, d4, d3 \n\t" // multiply result by scale
- "vshrn.i16 d0, q2, #8 \n\t" // shift down result by 8
- "vst1.32 {d0[0]}, [%[dst]] \n\t" // store result
- :
- : [x] "r" (x), [y] "r" (y), [a00] "r" (a00), [a01] "r" (a01), [a10] "r" (a10), [a11] "r" (a11), [dst] "r" (dst), [scale] "r" (scale)
- : "cc", "memory", "d0", "d1", "d3", "d4", "d5", "d6", "d7", "d16"
- );
+ vx = vdup_n_u16(x); // duplicate x into vx
+ vconst16_16 = vmov_n_u16(16); // set up constant in vconst16_16
+ v16_x = vsub_u16(vconst16_16, vx); // v16_x = 16-x
+
+ tmp = vmul_u16(vget_high_u16(tmp1), vx); // tmp = a01 * x
+ tmp = vmla_u16(tmp, vget_high_u16(tmp2), vx); // tmp += a11 * x
+ tmp = vmla_u16(tmp, vget_low_u16(tmp1), v16_x); // tmp += a00 * (16-x)
+ tmp = vmla_u16(tmp, vget_low_u16(tmp2), v16_x); // tmp += a10 * (16-x)
+
+ vscale = vdup_n_u16(scale); // duplicate scale
+ tmp = vshr_n_u16(tmp, 8); // shift down result by 8
+ tmp = vmul_u16(tmp, vscale); // multiply result by scale
+
+ vres = vshrn_n_u16(vcombine_u16(tmp, vcreate_u16(0)), 8); // shift down result by 8
+ vst1_lane_u32(dst, vreinterpret_u32_u8(vres), 0); // store result
}
